mock testa mmcfinalfinalreportmath.uaa.alaska.edu/~orca/mock_testa_mmcfinalreport.pdf ·...
TRANSCRIPT
1
Final Report: An Agent-Based Model of Predator-Prey Relationships
Between Transient Killer Whales and Other Marine Mammals Principal Investigators: Kenrick J. Mock and J. Ward Testa
Student Participants: Cameron Taylor, Heather Koyuk, Jessica Coyle, Russell
Waggoner, Kelly Newman
Date: May 31, 2007
Introduction
The role of killer whales (Orcinus orca) in the decline of various marine mammal
populations in Alaska is controversial and potentially important in their recovery.
Springer et al. (2003) hypothesized that declines in harbor seal, Steller sea lion and sea
otter populations in Alaska were driven by the over-harvest of great whales in the 1950’s
– 1970’s, leading to a cascade of prey-switching by killer whales from these large prey
species to smaller, less desirable prey species. That hypothesis is opposed by many
cetacean researchers , who cite inconsistencies in the timing of declines, insufficient
killer whale predation on large whales, and the absence of declines in other areas with the
same patterns of commercial whaling (DeMaster et al. 2006, Mizroch and Rice 2006,
Trites et al. 2007 In Press, Wade et al. 2007 In Press). Whatever the role of commercial
whaling, it is known that killer whales prey on threatened marine mammal populations in
the North Pacific (e.g., sea otters, Enhydra lutris, and Steller sea lions, Eumetopias
jubatus), and that the magnitude of that predation is at least a plausible factor either in
their decline or in their failure to recover (Estes et al. 1998, Heise et al. 2003, Springer et
al. 2003).
Estimation of killer whale numbers and the rates of predation on various marine
mammal species now have high research priority, but how we interpret these new data is
dependent on having an adequate theoretical framework. Thus far, only simplistic, static
models of killer whale consumption have been constructed to test the plausibility of killer
whale impact on other species (e.g., number of whales × killing rate of Steller sea lions =
estimated impact; Barrett-Lennard pers. comm.). Classical approaches to modeling
predator-prey relationships are rooted in the Lotka-Volterra equations. These entail
estimation of a “functional response” that defines the number of prey that can be captured
and consumed as a function of the densities in which they are encountered, and a
“numeric response” of the predator that describes the efficiency with which prey are
converted into predators. Data to support a form for either of these functions in transient
killer whales and their prey are essentially non-existent. Studies of diet in transient killer
whales are accumulating, but are rarely combined with information on prey-specific
abundance or availability. Moreover, there is little theoretical development for dynamic
predator-prey systems involving a single predator interacting with the number and
diversity of prey species hunted by transient killer whales, and less of a framework for
understanding how hunting in groups might affect even simple models.
While data to support development of a classical predator-prey model for killer
whales are sparse, research on the biology and behavior of killer whales as individuals
and groups has greatly accelerated in recent years. This suggests that one approach to
theoretical development of predator-prey models might be the implementation of
Individual-Based Models (IBMs) that use these recent studies to evaluate properties of
2
predator-prey relationships that emerge from our knowledge, and uncertainties, about the
biology of individuals and social groups of transient killer whales. At this point, we do
not know the relationship between killing rate of killer whales and prey densities
(functional response), nor the relationship between prey abundance or consumption and
population growth in killer whales (numeric response). However, we have some ideas
about the energetic requirements of these large predators, the size and structure of
hunting groups, and the number and kinds of prey pursued and killed in certain places
and times of the year. Following the guidelines proposed by Grimm and Railsback (2005)
(Grimm and Railsback 2005) for IBM modeling in ecology, we propose that our
knowledge at these levels can be used in an IBM to reproduce the characteristic emergent
patterns in group size, prey consumption and demographics of killer whales and their
prey, and to then explore how assumptions of such models influence more complex
emergent properties such as functional and numeric responses, or how depletion of
selected prey resources (e.g., removal of large whales by humans) might change predator-
prey dynamics under different assumptions. Perhaps more importantly, such an exercise
may also identify critical conceptual elements and critical real-world data essential to
understand the most obvious characteristics of the killer whale predator-prey system in
the NE Pacific Ocean…e.g., the persistence and basic population dynamics of transient
killer whales and their marine mammal prey.
Our objectives here are to reproduce characteristic patterns of demography, social
structure and prey consumption observed in transient killer whales by implementing
models of life history, energetics, and social associations at the level of individual killer
whales, and predator-prey interactions at the level of hunting groups of killer whales.
Recent studies of prey consumption and the structure of hunting groups (Baird and Dill
1995, Baird and Dill 1996, Baird and Whitehead 2000) of transient killer whales were
used to (1) formulate and parameterize the components of an agent-based model, and (2)
make comparisons to the emergent properties of these models as a form of model
validation. Detailed information on demography of transient killer whales is unavailable,
so we relied on comparisons to the demography of resident killer whales (Olesiuk et al.
1990) to arrive at similar vital rates and age-sex composition, and to patterns
characteristic of density-dependent changes in other large mammals (Gaillard et al. 1998,
Eberhardt 2002) when confronted with food shortages that have not been reported from
studies of transient killer whales thus far. Knowledge of killer whale energetics is sparse
(Kriete 1995, Williams et al. 2004), but we patterned our approach after that of Winship
et al. (Winship et al. 2002) for Steller sea lions with adjustments for the allometric
relationship suggested by Williams et al. (2004). We view this model as a first step
toward models that incorporate better formulations of any of these components, and
models with explicit movements and spatial structure. As such, it is a work in progress;
various upgrades and innovations in implementation are likely to be found when
consulting documentation and downloads at our website:
http://www.math.uaa.alaska.edu/~orca/ . The model components will be described below,
with additional details given on our website and in the Appendix. This report and links to
our downloads are available at www.mmc.gov.
3
Model Components
Individual Transient Killer Whales
Individual killer whale agents in this model possess characteristics allowing for
complete age and sex structure of the population to be “sampled” at a user-chosen day or
days of the year (usually summer, to correspond with most field research on transient
killer whales), as well as mass, reproductive status, and known maternal parent to
establish kinship along matrilines. Each whale therefore has:
Unique ID
Birthdate (and therefore age)
Sex
Mass
Reproductive status (pregnant, lactating)
Identity of mother (and therefore relationships to siblings and other relatives)
Group membership with other killer whales while hunting
Record of past associations with other whales
Baseline Demography
The model assumes underlying rates of birth and death that derive from causes
unrelated to rates of prey consumption, as distinct from those that are mediated by the
ability to maintain an expected body mass for that age and gender. These can be given as
baseline probabilities of becoming pregnant or dying (Fig. 1) that yield maximum rates of
growth with unlimited food. Olesiuk et al. (1991) suggest that the maximum rate of
growth in resident killer whales is around λ = 1.04, and default values for this model
(Table 1) are drawn from their life table to produce such growth when prey are abundant.
Age-Specific Survival & Conception
00.10.20.30.40.50.60.70.80.9
1
0 20 40 60 80Age (years)
Ann
ual P
roba
bili
ty
Female Survival
Male Survival
P(Conception)
Figure 1. Baseline annual probabilities of survival and conception for a
population of transient killer whales unlimited by prey availability.
Body Mass Dynamics
Individual Target Mass (TM)
Mass dynamics are based on a von Bertalanffy (von Bertalanffy 1938) growth curve
(Fig. 2) defining a gender and age-specific target mass (Table 1). Asymptotic weights
and growth rates were approximated from captive killer whales (Clarke et al. 2000).
4
Table 1. Parameters and files used in agent-based simulation model and controlled by the user.
Killer Whale Model Parameters Model Compartment Model Component Parameter or File Name Default Value
Execution
Day of year that model variables are sampled for output SampleDate 243
Starting files and conditions for model execution BatchFileName batch.txt
Run Length BatchRunLength 1
Demographic Rate File Fileparameters popparms.csv
Starting Population File FilePopulations population50.csv
To control diagnostic messages ShowDiagnostics FALSE
To suppress screen output BatchMode TRUE
Prey populations and vulnerabilities FilePrey prey.csv
Demographic Age-specific annual probabilities of conception popparms.csv
Age-specific annual probabilities of survival popparms.csv
Beginning age & sex structure, relatedness population50.csv
Conception date MeanDayPregnant 165
Conception date standard deviation StDevDayPregnant 35
Gestation length DaysPregnancy 510
Mass Dynamics
Von Bertalanffy asymptotic female mass FemaleMaxMass 2400
Von Bertalanffy growth exponent for females FemaleVonBert 0.0003
Von Bertalanffy asymptotic male mass MaleMaxMass 4000
Von Bertalanffy growth exponent for males MaleVonBert 0.0025
Proportion of target mass needed to maintain pregnancy AbortionThreshold 0.75
Mass of calf at birth BirthMass 182
Maternal mass gained, then lost at birth as proportion of calf mass PregnancyTissueMass 0.2
Proportion of target mass at which all lactation stops LactationCease 0.75
Proportion of target mass needed to maintain full milk production LactationDecrease 0.85
Extra mass gained during pregnancy to support future lactation PregnancyWeightGain 0
Proportion of target mass at which metabolism is reduced StarveBeginPercent 0.9
5
Proportion of target mass needed to avoid death by starvation StarveEndPercent 0.7
Fetal Growth BirthMass / (1 + e(a×(t+b)
a = -16, b = -0.68
Energetics
Efficiency of energy conversion into fetal growth EnergyToFetusEfficiency 0.2
Efficiency of energy conversion into tissue growth EnergyToMassEfficiency 0.6
Efficiency of energy conversion into milk EnergyToMilkEfficiency 0.75
Field Metabolic Rate Constant (kcals) FMRConstant 405.39
Field Metabolic Rate Exponent( kcals) FMRExponent 0.756
Maximum daily prey consumption as proportion of target mass GutMassPercent 0.055
Efficiency of tissue catabolism for maintenance energy MassToEnergyEfficiency 0.8
Energy content of milk (kcals/g) MilkKcalPerGram 3.69
Digestive efficiency of converting milk into energy MilkToEnergyEfficiency 0.95
Digestive efficiency of converting prey tissue into energy PreyToEnergyEfficiency 0.85
Caloric value of killer whale mass (kcals/kg) WhaleKcalPerKg 3408
Group Dynamics
Daily probability of meeting another group of killer whales for hunting ProbGroupsMeet 0.7
Daily probability that group is unrelated ProbJoinRandomGroup 0.1
Predator-Prey Prey population parameters (see text) Prey.csv User specified
Predator-prey interaction parameters (see text) Prey.csv User specified
Age killer whales reach full hunting effectiveness HuntAgeMax 12
Age juveniles begin to contribute to prey capture HuntAgeMin 3
Maintain constant annual prey population size for debugging UseConstantPreyPopulation false
Starting population of juvenile prey n_0 prey-dependent
Starting population of non-juvenile "adult" prey n_adult prey-dependent
Day of prey's annual birth pulse BirthDate prey-dependent
Mass of juveniles at birth n0_startmass prey-dependent
Mass of juveniles after 1 year n0_endmass prey-dependent
Mean mass of adult prey ad_mass prey-dependent
6
Caloric value of juvenile prey n0_kcals_gram prey-dependent
Caloric value of adult prey ad_kcals_gram prey-dependent
Maximum birth rate of adults (>1 year) BirthMax prey-dependent
density dependent birth parameter a in exp(-a * N^b) Birth_a prey-dependent
density dependent birth parameter b in exp(-a * N^b) Birth_b prey-dependent
Maximum juvenile survival n0Surv_Max prey-dependent
density dependent juvenile survival parameter a in exp(-a * N^b) n0Surv_a prey-dependent
density dependent juvenile survival parameter b in exp(-a * N^b) n0Surv_b prey-dependent
maximum adult survival AdSurv_Max prey-dependent
density dependent adult survival parameter a in exp(-a * N^b) AdSurv_a prey-dependent
density dependent adult survival parameter b in exp(-a * N^b) AdSurv_b prey-dependent
probability of encounter between killer whale group and juvenile prey 0_encounter_rate prey-dependent
maximum vulnerability of juvenile prey to large killer whale groups 0_VulnMax prey-dependent
logistic parameter a for group-dependent vulnerability of juveniles 0_VulnA prey-dependent
logistic parameter a for group-dependent vulnerability of juveniles 0_VulnB prey-dependent
probability of encounter between killer whale group and adult prey ad_encounter_rate prey-dependent
maximum vulnerability of adults to large killer whale groups ad_VulnMax prey-dependent
logistic parameter a for group-dependent vulnerability of adults ad_VulnA prey-dependent
logistic parameter b for group-dependent vulnerability of adults ad_VulnB prey-dependent
day of year prey become available to killer whales Available_Start prey-dependent
day of year prey become unavailable to killer whales Available_End prey-dependent
7
Killer Whale Growth
0
1000
2000
3000
4000
5000
0 10 20 30 40
Age (years)
Mas
s (
kg
)
.
Males
Females
Figure 2. Von Bertalanffy growth model of age-specific target mass for
transient killer whales.
Gestation
Growth of the fetus and associated maternal tissues is considered additional to the
normal age-specific mass of a female calculated in Fig. 3. A general fetal growth model
was used (Winship et al. 2002):
Fetal Mass = (BirthMass) / (1 + e(a×(t+b))
),
where t is proportion of total gestation length (510 days, BirthMass=182, a = -15 and
b = -0.68.
0
50
100
150
200
0 100 200 300 400 500
Days Pregnant
Fe
tus
Mas
s (
kg
)
Figure 3. Fetal growth in model killer whales.
It is assumed that the pregnant female supports an additional mass
(BirthMassLoss=0.2) proportional to fetus mass for placenta and blood that must be
grown during pregnancy, but is lost from her actual mass and Target Mass (TM) at birth.
8
An additional parameter (PregnancyTissueMass) is allowed for mass gain that may occur
in preparation for lactation following birth, but it is unknown if killer whales actually
store energy for this purpose and the default setting is 0.
Regulation of Body Mass
Body mass is regulated by reducing the amount of food consumed when an
individual killer whale approaches or exceeds its age and sex-specific target mass. Our
model assumes that a killer whale’s maximum daily consumption (GutMassPercent) is a
fixed proportion of its Target Mass as computed by the von Bertalanffy growth curve. If
an animal is underweight, we expect it would attempt to eat an amount near this
maximum, and if very fat would eat only as much as it takes to meet its daily metabolic
requirements, including those for gestation and lactation demands. We estimated the
proportion of a whale’s maximum daily consumption that would be required to meet
daily metabolic requirements and used the remainder to estimate the remaining gutfill
that could be used to fuel body growth. We used a logistic function to describe the
proportion of remaining GutMassPercent that an animal would attempt to consume (i.e.,
beyond its metabolic needs) in relation to its actual mass/target body mass (Fig. 4). The
mass of food required to meet this satiation level is based on the energy content of a
preferred prey (harbor seals), rather than the energetic content of the diet on that
particular day.
0
0.2
0.4
0.6
0.8
1
0.70 0.90 1.10
Actual/Target Body Mass
Re
ma
inin
g G
utfill
Figure 4. Proportion of remaining stomach volume (beyond that needed
for maintenance metabolism) that will satiate an individual killer whale in
relation to body condition (actual mass/target mass).
Killer whale calves transition gradually from milk to prey that are killed by its
mother or other members of its pod or hunting group, probably within their first year
(Heyning 1988). We assumed a logistic model (a=6.1, b=-0.02) that reduces the
proportion of milk in a calf’s diet gradually (Fig. 5). The energetic needs of the calf and
food volume required for satiation were calculated by the same metabolic formula
described for adults, with higher metabolism generated by the exponent of the field
metabolic rate (FMR) and by requirements for body growth. The proportion of that
target that was milk was used to calculate the energetic demand on the female as part of
9
her daily energy requirement, and if she could provide it the calf’s diet included that
energy. The remainder of the desired amount of food for the calf came from prey
captured by the calf’s hunting group, if available.
0.00
0.20
0.40
0.60
0.80
1.00
0 200 400 600 800
Calf Age (days)
Pro
port
ion
Milk
Prey
Figure 5. Logistic model of calf diet describing age-related transition from
milk to prey.
Thresholds
The growth and consumption models described above produce individuals with
variation in realized mass around that predicted from the age and sex specific growth
curve, much as we see in natural populations. The model uses realized individual body
mass to impose demographic consequences (e.g., births, deaths, aborted pregnancies or
termination of lactation) when the killer whale fails to maintain its mass above user-
specified thresholds of its target mass (Moen et al. 1997, Moen et al. 1998). The ability to
maintain body mass is determined by the energetic requirements of the killer whales and
their prey consumption. The parameters controlling thresholds (Table 1) are expressed as
proportions of the age specific target mass of a whale, and can be modified at the start of
a simulation. Our default assumptions are that whales begin to starve at 0.85 of their
target mass and their field metabolic rate declines to half normal in a linear fashion until
starvation occurs at 0.7 of their target mass. Similarly, milk production by lactating
females is reduced linearly from its normal value to 0 as the female’s mass falls from
0.85 to 0.75 of its target mass (Table 1). Tissues associated with gestation (fetus and
maternal tissue) are considered part of the female’s TM additional to that calculated from
her age-specific growth curve (Fig. 2) when setting mass-dependent satiation (but not
GutMassPercent) levels.
Energetics
The requirements and efficiencies of converting prey or body mass into energy,
and using that energy to support field metabolic rate (FMR) or somatic production
(Fig. 5) are similar to those used by Winship et al. (2002) for Steller sea lions
(Eumetopias jubatus). We make the simplifying assumption of a constant ratio of lean to
fat tissue in the body of killer whales with an average energetic value of killer whale
tissue of 3.4 kcals/g. Given that the metabolic rate of lean probably exceeds that of fat
10
tissue, this may underestimate the metabolism of starving whales and overestimate that of
well conditioned whales, but this was considered an acceptable cost for simplifying the
model, and its effect could be compensated for by adjustments in threshold values.
Moreover, actual metabolic rates of starving and well-conditioned whales are unknown.
EnergeticsPrey
2.5-3.5 kcals/g
Killer WhaleGutfill <= 4.5% body mass
85% assimilationCatabolism 60% effic.
3.4 kcals/g
MetabolismFMR = 405*Mass^0.756FMR/activity declines
near starvation
Body Growth60% effic.
Energy
CalfGutfill 4.5%95% effic.
Milk3.69 kcals/g75% effic.
Fetus20% effic.
Simplifying Assumption: constant body composition…lean tissue dominates equations, uncertainties much greater in other components
Figure 6. Energetic model by which prey are converted into energy for
metabolism, body growth and reproduction of individual killer whales.
The energetics of transient killer whales are based on the estimates of field
metabolic rate (FMR) for delphinids (Williams et al. 2005):
FMR = 405.39 × M0.756
kcals/day,
where M = mass in kg. Metabolic and catabolic conversion efficiencies are similar to
those suggested by Winship et al. (2002: Fig. 6 & Table 1), though few are based on
killer whale studies and many are poorly known or unknown in any marine mammal. The
requirements for fetal growth and lactation, including the efficiencies in Fig. 6, are added
to the female’s FMR when determining the daily energetic maintenance requirements.
The higher mass-specific energy generally needed by juveniles (Winship et al. 2002) is
accounted for by explicitly modeling somatic growth and by the allometric
parameterization of FMR (Williams et al. 2005).
Group Dynamics
We modeled the self-formation of groups based upon rules for aggregation and
dispersal to optimize a fitness function that explores the tradeoff between individual and
group optimality (Aviles et al. 2002, Parrish et al. 2002). Because there is no spatial
component that could be used to generate “encounters” between groups, these are
EnergeticsEnergeticsPrey
2.5-3.5 kcals/g
Killer WhaleGutfill <= 5.5% body mass
85% assimilationCatabolism 60% effic.
3.4 kcals/g
MetabolismFMR = 405*Mass^0.756FMR/activity declines
near starvation
Body Growth60% eff.
Energy
CalfGutfill 5.5%95% effic.
Milk75% eff.
Fetus20% effic.
Simplifying Assumption: constant body composition…lean tissue dominates equations, uncertainties much greater in other components
11
generated probabilistically, with weighting toward groups that have a history of
associations, such as near relatives. Our model allows approximate optimization of group
size by maximizing the expected amount of prey each individual can expect to eat in a
group while incorporating the effect of familial bonds that constrain the possible choices
of hunting partners.
The Maternal Unit
Our model for social aggregation into hunting groups is based primarily on the
mother-calf bond, which probably persists for female calves until they begin to reproduce
and nearly indefinitely for male offspring unless an older brother is already present
(Baird and Whitehead 2000). Dispersal of females occurs with the birth of their first calf
(see demographics for age of first reproduction). For males in groups with an older male
sibling already resident, dispersal occurs at sexual maturity, which is currently set at 12.
Histories of Association
Each model killer whale maintains a “memory” of its past associations with all
other killer whales. It is this history that determines the probability of associating with a
whale that is not its mother in the future, rather than relatedness per se. The effect is that
siblings will tend to associate with their mother and with other siblings even after
dispersal, but that those associations will be weaker with larger discrepancies in age.
The association memory is implemented by incrementing counters for all whales
in a group during the daily time step. For example, consider two groups of whales shown
below. Group 1 consists of two whales with ID’s #1 and #2. Group 2 consists of one
whale with ID #3. In group 1, whale #1 and whale #2 have been in the same group for
150 time steps. Whale #1 and whale #3 were previously in the same group for 20 time
steps, although both whales are currently in different groups.
If whale #1 has a newborn calf then in the next time step a counter for the calf
will be added for all other whales in the group. Additionally, the counters are
incremented for all whales in the group. This is shown below where the newborn is
whale #4.
If the calf is in the same exact group the next time step then those counters will be
incremented to 2. If at some point in the future a new whale joins the calf’s group then a
Whale ID Counters
1 3→20, 2→151, 4→1
2 1→151, 4→1
4 1→1, 2→1
Group 1
Whale ID Counters
1 3→20, 2→150
2 1→150
Group 1
Whale ID Counters
3 1→ 20
Group 2
12
similar suite of new counters will be created for that whale that are initialized to 1. When
a whale dies the counters are removed. In this manner, each whale maintains a count of
how frequently it has associated with other whales, which forms the basis from which
whales can organize into hunting groups. A majority of these associations will be due to
familial relationships.
Hunting Groups
The grouping behavior of the whales affects vulnerability of prey and ability to
hunt certain types of prey. Field research indicates that 3 whales may be the optimal
group size for medium-sized mammals like harbor seals or harbor porpoises, while larger
groups may be more effective for hunting gray whale calves (Baird and Dill 1996).
Association of maternal groups with more extended family members is sometimes
observed when transients are hunting, and is likely related to the effectiveness of larger
groups for certain types of prey, such as whales or large pinnipeds. We assume that there
is an optimum group size for hunting each type of prey available to killer whales
(described in section on Predator-prey Interactions), and that the optimum group size at
any time depends on the numbers of each prey type available.
We have implemented a model for groupings larger than the basic family unit by
allowing smaller groups to combine together. As modeled here, the probability of a
group of whales interacting with a different group each day is controlled by two
stochastic variables chosen by the user: ProbGroupsMeet for the probability that a group
of whales will meet another group of whales during the time step, and
ProbJoinRandomGroup for the probability that the group encountered is an arbitrary
group of whales that may or may not have been associated together in the past. A
uniform random number generator is used to generate these encounters. The number is
generated per group, so it is possible for some groups to join and others to maintain their
existing group structure during one time step. Note that ProbGroupsMeet is applied
before ProbJoinRandomGroup. Only after it is determined that groups will meet is the
decision made whether the group will be arbitrary or based on association histories.
If two groups are meeting based on association histories then this group is chosen
randomly with a weight proportional to the number of past associations of all group
members. An example is shown in Figure 7. Here we are trying to determine if group 3
should meet with group 1 or group 2. Whales from group 3 have interacted a total of 60
times with whales from group 1 (whale #4 twenty times with whale #1, whale #4 ten
times with whale #2, and whale #5 thirty times with whale #2). Similarly, the whales
from group 3 have associated a total of 50 times with whales from group 2. As a result,
group 3 will meet group 1 with probability (60/110) and it will meet group 2 with
probability (50/110).
13
Figure 7. Determining probabilities for group encounters.
More formally, we compute the probability P(gx,gy) of group gx encountering
group gy where whale wi refers to a whale within a group we use:
( )
( )∑
∑ ∑
−∈
∈ ∈
=
=
xi
xx yy
gAllGroupsg
ix
yx
yx
gw gw
yxyx
ggWeight
ggWeightggP
wwtionsNumAssociaggWeight
),(
),(,
,),(
Two exceptions to these calculations are groups with mature males that have left
their mother’s group due to an older sibling or females that have left their mother’s group
due to the birth of a calf. The dispersal rules prevent these whales from joining their
mother’s group and in these cases the mother’s group is removed from the calculations.
When two groups meet they do not automatically join together. Only after two
groups of whales have been selected that satisfy the encounter conditions do we evaluate
whether or not the two groups will join together. Larger groups can more effectively
hunt larger prey, but captured prey must now be shared among all group members. To
optimize these competing factors the model uses the larger of the two groups to
determine the outcome by computing the expected amount of food per individual based
on the vulnerability of the prey as a function of group size (see Group-size Dependent
Prey Vulnerability below). The list of prey used in this calculation is the actual prey that
the group has encountered in the simulation the previous day, as opposed to the true
number of prey that exists globally in the simulation. If this “energy” value is larger in
14
the combined group than the original group then the two groups join together. Otherwise,
no join occurs even if the smaller group might experience a larger energy gain by joining
the larger group. This amounts to an assumption of optimal foraging for the larger group,
with constraints imposed by the size of the groups interacting (e.g., 2 groups of 3 can
only form a group of 6, or remain separate on the day of their encounter).
In addition to accretion a group will also consider whether or not it is
advantageous to split into sub-groups on a daily basis. In this operation the largest sub-
group (what was once an original group that joined to form a larger group) computes
whether the energy value will be optimized by remaining in the larger group or by
splitting into its own group and selects the optimal choice. Conditions that may lead to
this scenario include the death of whale(s), a change in the prey encountered, or a change
in the group’s composition based on the rules described in the section on the Maternal
Unit.
Predator-prey interactions
Density-dependent Prey Populations
Models of the prey populations were constructed to be as simple as possible while
incorporating features considered essential, both from the standpoint of allowing different
vulnerability of juveniles and adults, and of incorporating realistic potential for density-
dependent population productivity. We considered the following elements to be essential
to our prey populations:
• Density-dependent growth rates of marine mammals is non-linear, with maximum
productivity declining rapidly near equilibrium (Fowler 1981, Eberhardt and
Siniff 1977, Eberhardt 2002).
• The magnitude of density-dependent changes is likely to be greatest in juvenile
survival followed by adult reproduction, and be least in adult survival (Gaillard et
al. 1998, Eberhardt 2002).
• Many prey species, including whales and large pinnipeds, are more vulnerable to
predation by killer whales in their first year of life than as older animals (Heise et
al. 2003, Wade et al. 2007 In Press)).
All prey populations were modeled as 2 age-classes: “age 0 years” and “adults”,
with 3 density-dependent (DD) vital rates: a survival rate for each age class and per
capita birth rate for the adult class. A Ricker function with 2 parameters (a & b) was used
for all 3 rates as a function of total prey population size, N:
Rate = Max Rate × exp(-a * Nb).
All parameters in the model are defined at the annual rate, so that difference equation
models on a 1 year time step could be used to generate plausible values and validate
outputs; the 365th
root of the calculated survival was used to model daily survival
proportions while the birth rate was applied on the species-specific birthing day annually.
Maximum survival and birth rates were chosen to produce maximum population growth
rates (λ) typical of particular species and life histories (e.g., ~1.12 for pinnipeds and small
cetaceans, 1.08 observed of humpback whales) with adjustments to compensate for the
fact that full age-sex structures were not being used (e.g., less than observed adult birth
rates to account for pre-reproductive ages being included in the “adult” model class).
Similarly, density dependent parameters were chosen to produce the general pattern of
15
maximum productivity at 70-75% of equilibrium, and the greatest magnitude of changes
in juvenile survival, birth rates and adult survival, in that order (Fig. 8). All the
parameters, and prey populations used in the model are user-controlled, and developed in
an interactive spreadsheet (PreyWorksheets.xls available in download package online).
Default values for a complex prey-field were derived to be consistent with published
accounts of 11 prey populations important to the stock of transient killer whales known to
inhabit the coastline from California to SE Alaska (Appendix). However, most
simulations to date were conducted using a single, or few prey species with parameters
generating much larger populations of prey in order to compensate for the absence of
those alternative prey populations. These simpler models were used to assess whether the
model was producing realistic population-level behavior of killer whales under conditions
of abundant or limiting prey, and to compare the dynamics to classical models of a single
predator and single prey species. Changes in prey vital rates and density dependence to
simulate “regime shifts”, or extraneous “removals” of known numbers to simulate human
harvest can be input as options during execution.
0.00
0.20
0.40
0.60
0.80
1.00
0 100,000 200,000
Population Size
Rate
BirthRateCalfSurvAdSurv
02000400060008000
1000012000140001600018000
0 100,000 200,000
Population Size
Net P
roduction
Figure 8. General density-dependent properties of vital rates (left) and net
productivity (right) of model prey populations.
Group-size Dependent Prey Vulnerability
While relatively little is know about vulnerability of prey with age, greater
vulnerability of juveniles is a common feature of predator-prey interactions, particularly
as the size of the prey species relative to that of the predator becomes larger. In the case
of transient killer whales, vulnerability of large whales is largely limited to calves(Wade
et al. 2007 In Press), and there also appears to be greater vulnerability of Steller sea lion
pups in comparison to older animals (Heise et al 2003). This was considered an essential
element to the prey model, while finer distinctions of sex and age were ignored. We also
assumed that larger groups of killer whales would be more effective at killing prey,
especially large prey, but the effect of sharing the prey in larger groups would produce an
optimum group size for each prey type that produced the greatest amount of prey biomass
per individual in the group (Baird and Whitehead 2000).
We implemented a model of killing rate similar to a classical formulation of
attack rate × number of prey, with attack rate partitioned into an encounter rate (e)
defined as the probability that a group of killer whales would encounter a particular
individual prey, and vulnerability (v) equal to the probability of being killed by the group
once encountered (i.e., expected kills per day equals e × v × number of prey available).
16
To make this dependent on group size (x), we used a simple logistic function (e.g., Fig. 9)
with a user-defined maximum vulnerability and logistic parameters a and b:
v = vmax * exp(a + b × x) / (1 + exp(a + b × x))
The logistic function was chosen for its generality and congruence with potential analyses
of field data. Calf and juvenile killer whales are not as effective hunters as adults, so
group size for this purpose was considered to be “adult equivalents”, where juveniles
began a linear increase in hunting effectiveness at age 3 (HuntAgeMin equivalent to 0
adults) and were considered fully effective hunters at age 12 (HuntAgeMax equivalent to
a single adult). Thus, a group of killer whales comprised of animals aged 1.5, 7.5, 24.5,
36.5 and 60.5 years would have an effective group size of 3.5 for hunting purposes. We
also linearly reduced the effectiveness of whales that become malnourished from full
effectiveness to 0 effectiveness as metabolic rate declines (BeginStarve = 0.85 to
EndStarve = 0.7, see section on Energetics). Thus, a group of two adult killer whales
where one is at 0.95 of target mass and the other is at 0.75 of target mass would have an
effective group size of 1.33. In this way each age class of each prey species could be
assigned plausible maximum vulnerabilities when encountered by a large group of killer
whales, and differences in vulnerability with hunting group size could be modeled with a
simple form that produces optimal predictions of individual gain per kill. Fig. 9 shows
this relationship for a small prey species such as harbor seals, while Fig. 10 shows a
similar relationship for a large species class, such as gray whale calves. When adjusted
for the size and energy value of particular prey and summed over all prey types available,
the expected optimum group size for any suite of prey abundances can be calculated (and
employed in choosing group sizes, as described earlier). This assumes no foraging
specialization by killer whale groups, which we consider a reasonable default assumption
that might be studied later.
0.0
0.2
0.4
0.6
0.8
1.0
1.2
0 5 10 15 20Group Size
Vuln
era
bili
ty
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0 5 10 15 20Group Size
Kill
s/K
ille
rWhale
Figure 9. The left graph shows the modeled relationship between the
vulnerability (probability of being killed given an encounter with a group
of transient killer whales) of a vulnerable prey type (e.g., harbor seals) and
the effective size (adult equivalents) of the hunting group, while the right
graph gives the resulting expectations of kills available as food per killer
whale from each encounter.
17
0.00.10.20.30.40.50.60.70.80.91.0
0 5 10 15 20Group Size
Vuln
era
bili
ty
0.00
0.01
0.02
0.03
0.04
0.05
0 5 10 15 20Group Size
Kill
s/K
illerW
ha
le
Figure 10. The left graph shows the modeled relationship between the
vulnerability (probability of being killed given an encounter with a group
of transient killer whales) of a difficult-to-kill prey species (e.g., gray
whale) and the effective size (adult equivalents) of the hunting group,
while the right graph shows the resulting expectations of kills available as
food per killer whale from each encounter.
Prey Capture
In executing a daily time step of foraging for a killer whale group, the model steps
through all prey types to determine the number of prey encountered of each type, drawing
random variables from a Poisson distribution with expectations equal to ei × number (ni)
of prey type i. Once all prey encounters are identified, their order is randomized and each
is subjected to a random trial to see if the encountered prey is killed by comparing its
vulnerability (e.g., Figs. 9 & 10) to a uniform random variable. The group kills prey in
the list until the list is exhausted or enough prey are consumed to sate all the individuals
in the group (Fig. 3). The kills are shared proportional to the mass required for each killer
whale in the group to satisfy its maintenance metabolic requirements and reach satiation.
Model Execution and Output
The model is executed in daily time steps as illustrated in Fig. 11. At the
beginning of each simulated day whales hunt and feed. Once all feeding has occurred,
metabolism and growth algorithms are applied, and demographic actions are taken.
Finally, any changes in group membership for the following day are determined. Various
flags are set to mark annually occurring events such as birthing and sampling for model
output (Table 1). Running annual totals are kept of births, deaths, and prey consumption
by killer whale age and sex class. Graphical output is provided during interactive
computer runs, but practical running times are obtained only in batch mode, where
pre-programmed commands control program variables and output files that are analyzed
after execution is complete.
The model is written in Repast, a Java-based software package for agent-based
modeling (North et al. 2006). Output data are compiled on user specified sampling dates
and written to spreadsheet files for post-processing in spreadsheet or statistical software
(e.g., Microsoft Excel). Instructions for downloading the software and running the model
are at http://www.math.uaa.alaska.edu/~orca/ .
18
Figure 11. Model execution of daily routines is illustrated with a flow
diagram. Object shapes denote actions taken at either group, individual or
model level.
19
Sensitivity to Input Parameters
The conditions for these simulations were set to the simplest scenario of a single
prey population (harbor seals) sufficiently large to support an average population of 200-
300 killer whales (to minimize the effects of demographic stochasticity). Williams et al.
(2004) estimated that the average food requirements for a population of transient killer
whales would be equivalent to roughly a harbor seal/day/killer whale. If we assume that
the maximum net productivity of harbor seals occurs at 70% of equilibrium (K), and that
the growth rate is roughly 80% of maximum (λ~1.10), a population of 200 transient killer
whales would require at least 73,000 harbor seals/year. For this to be sustainable with the
assumed vital rates, a harbor seal population of at least 730,000 and a density-dependent
equilibrium in excess of 1,000,000 would be required. Our single-prey test simulations
therefore used a population of harbor seals that would equilibrate at ~ 1,400,000 in the
absence of predation, with vital rates, density dependence, and predator encounter and
vulnerability given in prey1.csv (in download package online). The model was allowed to
run until both predators and prey experienced periods of growth, decline and relative
stability. Sensitivity of relevant model output variables to variation across plausible
ranges of parameters was evaluated graphically while holding other parameters constant
during multiple runs of 1000 years.
In general, the parameter space that allowed both species to persist was narrow.
Many of the parameters chosen for inclusion in the model have values or likely ranges
that can be supported with field data and do not generally lead to extinction of the
predators. However, some parameters are poorly known, and the plausible ranges may
greatly exceed the narrow parameter space that allows both species to persist. In classical
predator-prey models the attack rate, modeled here as the product of encounter rate and
vulnerability, greatly affects the persistence of a single predator-single prey system
(Metzgar and Boyd 1988); low attack rates lead to steady decline and extinction of
predators while high rates lead to oscillations and the extinction of one or both species.
Attack rates of transient killer whales in SE Alaska (Dahlheim and White, Pers. comm.)
suggest that encounter rates (probability that a particular group of killer whales would
encounter a single individual prey) might be on the order of 10-4
to 10-6
. For models with
a single super-abundant prey, encounter rates producing relatively stable killer whale
populations comprised a narrow range (Fig. 12) within the range plausible. For a set of
fixed parameters, encounter rates of <3.00E-06 led to rapid extinction of the killer
whales, while increasing the encounter rate above this threshold lead to increasing
numbers of killer whales, but eventually also to oscillatory behavior above 4.00E-06 (Fig.
12). The actual values needed to produce this progression varied with the choice of other
values for parameters (e.g., greater energetic efficiency could lower the values of
encounter rates needed for stability, or lower energetic efficiency could lead to
extinctions), but a narrow range of encounter rates needed for relatively stable numbers
of killer whales was characteristic of all simulations. The low number of reproductive
killer whales and simplistic assumptions about random encounters undoubtedly
contribute to model instability as encounter rates increase. Nevertheless, as a first
approximation of how killer whales might interact with prey, we proceeded by
20
0
100
200
300
400
500
0 200 400 600 800 1000
Year
Kill
er
Whale
s
Enc. = 3.0E-6
Enc. = 3.6E-6
Enc. = 4.4E-6
0
50
100
150
200
250
300
350
400
2.0E-06 4.0E-06 6.0E-06 8.0E-06 1.0E-05Encounter Rate
Kill
er
Wh
ale
s
0.95
1.00
1.05
Gro
wth
Ra
te (λ)
Mean Population SizeInitial Growth Rate
Figure 12. Three trajectories are shown (left) for different assumed
encounter rates between hunting groups of killer whales and individual
prey. The initial (maximum) growth rate of the killer whale population in
the first 50 years of simulations, and the mean abundance of killer whales
(SD in error bars) after 200 years of 5, 1000-year simulations in relation to
encounter rates (right) illustrates the effects of increasing predatory
efficiency on basic killer whale population dynamics.
determining a range of encounter rates that allowed persistence and relative stability of
killer whales as a necessary precondition for assessing other model parameters.
Many of the remaining parameters in the model are expected to have redundant
effects on model performance. For example, energetic efficiencies are part of a chain of
conversions, any one of which could be used to lower or raise the overall efficiency with
which prey are converted into predator biomass. Similarly, several thresholds set for
killer whale mass relative to target mass influence birth and age-specific survival rates,
and are therefore likely to affect population recruitment, survival and growth. Our
implementation includes some redundancy in function in relation to the objective of
simulating predator-prey dynamics. We therefore tested sensitivity of certain model
output variables in relation to plausible variation in model parameters, but ignored
parameters whose action duplicated the effect of others (e.g., energetic efficiency of milk
production, milk energy content, and calf digestive efficiency all have similar effects on
energy transfer from mother to calf) or obviously had small influence. (e.g., birth mass
alters both fetal and calf growth trajectories, but with little expected influence on
mortality, so it was ignored). We focused on (1) thresholds of body mass (as a proportion
of desired target mass) for abortion, cessation of lactation, and starvation, (2) gut size as a
constraint to daily consumption and (3) energetic efficiencies of producing milk and
digesting prey. Each of the parameters was evaluated graphically for its effect on mean
number and standard deviation of killer whales after 200 years of growth from identical
starting conditions in simulations of 1000 years for each value of the parameter. Where a
likely demographic mechanism for the effect of a changing parameter value was apparent
(e.g., decreasing efficiency of milk production is expected to decrease neonate survival)
the mean value of the appropriate vital rate was also graphed against the value of the
parameter. The results are summarized in Figure 13 for 6 parameters.
The threshold body mass (as a proportion of an individual’s age-specific target
mass) that caused a pregnancy to be aborted was a sensitive parameter, causing little
effect when set near the starvation threshold (0.7-0.72), but it led to a lowered calving
21
0
100
200
300
400
0.68 0.72 0.76 0.80Abortion Threshold
Kill
er
Whale
s
0
0.1
0.2
0.3
0.4
Ca
lvin
g R
ate
Killer Whales
Calving Rate
0
100
200
300
400
0.70 0.74 0.78 0.82Lactation Threshold
Kill
er
Whale
s
0.00
0.20
0.40
0.60
0.80
1.00
Neonate
Surv
ivalKiller Whales
Neonate Survival
0
100
200
300
400
0.65 0.70 0.75
Death Threshold
Kill
er
Wh
ale
s
0.96
0.98
1.00
1.02
1.04
Ma
xim
um
Gro
wth
Killer Whales
Max Growth
0
100
200
300
400
0.00 0.05 0.10
Daily Consumption Limit
Kill
er
Wh
ale
s0.96
0.98
1.00
1.02
1.04
Ma
xim
um
Gro
wth
Killer Whales
Max Growth
0
100
200
300
400
0.0 0.5 1.0
Energy To Milk Efficiency
Kill
er
Wh
ale
s
0.00
0.20
0.40
0.60
0.80
1.00
Gro
wth
and
Su
rviv
al
Ra
tes
Killer WhalesMax GrowthNeonate Survival
0
100
200
300
400
0.60 0.70 0.80 0.90 1.00Prey To Energy Efficiency
Kill
er
Whale
s
0.96
0.98
1.00
1.02
1.04
Maxim
um
Gro
wth
Killer Whales
Max Growth
Figure 13. Sensitivity of killer whale numbers in the final 800 years of
1000-year simulations, maximum population growth in the initial 50 years, and
other key output variables are shown as key input parameters were varied
(5 replicates each, error bars = SD). Thresholds at which killer whales aborted
pregnancies (top left), stopped lactation (top right), and died (center left) are
expressed as the proportion of age-specific target mass. Daily consumption
limit (center right) is also a proportion of age-specific killer whale target mass.
Efficiency of milk production and the efficiency with which prey were
converted to energy are shown at bottom left and right, respectively.
rate at 0.74, and rapid extinction of the killer whales by 0.80 of target mass. The
threshold for cessation of lactation, and therefore death of neonates was also influential,
with a nearly linear effect on neonate survival from just above that causing death of the
mother (0.70) to complete mortality of neonates and extinction of killer whales at a
threshold of 0.80. The default threshold was selected to be 0.75, which produced neonate
mortality from birth to the first summer similar to that reported by Olesiuk et al. (1991)
while making it responsive to variable adult consumption rate in the models. The
22
starvation threshold less than the default of 0.7 had little effect on equilibrium killer
whale numbers, but higher thresholds led to lower growth rate, lower numbers and
extinction by a threshold of 0.74.
Energetic efficiencies modeled were essentially part of a conversion chain
producing predator biomass from prey biomass. The metabolic efficiency of converting
prey to energy (Fig. 13, bottom right) encapsulates the effect of any link in that energetic
chain and demonstrates a strong effect on total number of whales that could be sustained
and on the stability of whale numbers as efficiency was increased. It is very similar in
functional form to that of encounter rates (Fig. 12), The particular functional shape
illustrated in the figure can be shifted in either horizontal direction by changing some
other parameter affecting the metabolic or hunting efficiency of the killer whales, but the
progression from smaller to larger populations and toward more unstable population
trajectories with increasing efficiency was consistent. The results shown should not be
interpreted as supporting the assumption of a particular metabolic efficiency.
Group Size
We first evaluated whether model killer whale groups approximated the expected
optimum group size when we altered the shape of functions relating prey vulnerability to
group size. We chose parameters to create modal values of 2, 3, 4 and 5 for the test case
of superabundant harbor seals (Fig. 14). The optimizing models were based on “adult
equivalents” (downgrading juveniles for their expected lesser effectiveness) while the
model output included all individuals, so we expected the observed group size might
exceed the modal optimum, but group sizes are constrained by the available groups of
related whales with which each group might join. The logistic functions relating killer
whale group size to maximum vulnerability of prey produced shapes of the killing rate
per killer whale that were unimodal at the optimum group size (Fig. 14, upper graphs).
Adjusting the shapes of the vulnerability and killing rate curves altered the overall killing
rates and resulting equilibrium population size of killer whales, so the simulations were
standardized (by altering the encounter rates) to produce killer whale populations that
varied around 200 whales in the last 800 years of 1000-year runs (Fig. 14, bottom left).
The mean group size was close to the modal optimum size expected in the 4 test cases,
and the histogram of daily observed group sizes for an optimum group size of 3 (Fig. 14,
bottom right) was plausible when compared to those reported by Baird and Dill (1996).
Mean Group size was larger by 0.2-0.9 whales during the initial population growth phase
of the simulations, but this effect incorporates complex relationships of mother-offspring
association rules and the skewness of the per capita consumption to group size curves, so
was not quantified in a precise way.
Parameter ProbGroupsMeet sets the probability of encountering a group of killer
whales and evaluating the foraging advantage of combining with that group. The group
encountered is weighted by known previous associations (the more past associations, the
higher the chance of considering that group as hunting partners the following day). The
parameter ProbJoinRandomGroup sets the probability that the group encountered and
considered for partnership is a random group irrespective of past associations. This was
considered a plausible, but unlikely possibility based on literature accounts (Baird and
Dill 1996, Baird and Whitehead 2000). A randomly selected group would be more likely
to reflect the distribution of group sizes in the entire killer whale population, while the
choices among those groups previously known would be more limited, reflecting the
23
0.0
0.2
0.4
0.6
0.8
1.0
0 5 10Group Size
Vuln
era
bili
ty
0.00.10.20.30.40.50.60.70.80.91.0
0 5 10Group Size
Kill
s/W
ha
le
0
50
100
150
200
250
300
350
400
1 2 3 4 5 6Modal Optimum Group Size
Kill
er
Whale
s
2
3
4
5
6
Mea
n G
roup S
ize
Killer Whales
Group Size
0%
5%
10%
15%
20%
25%
30%
1 2 3 4 5 6 7 8 9 10
Group Size
Pe
rce
nt
Fre
que
ncy
Figure 14. The logistic functions relating killer whale group size to a
maximum vulnerability (probability of killing given an encounter) of prey
(top left) produce expected payoffs in consumable prey per group member
(top right, scaled to the maximum value with colors to match curves at top
left). Simulations of 1000 years with encounter rates scaled to produce
roughly the same numbers of killer whales in the last 800 years of each
run demonstrated that mean group size approximates the modal optimum
group size (bottom left). A Histogram of group sizes for a modal optimum
of 3 killer whales demonstrates the variability in daily group size (bottom
right).
number of living relatives, their reproductive success and their own particular hunting
associations. We tested the sensitivity of group size to variation in ProbGroupsMeet
while ProbJoinRandomGroup was held to 0, and varied ProbJoinRandomGroup while
ProbGroupsMeet was held at 1 (ProbJoinRandomGroup only operates after a simulated
encounter occurs, which depends on ProbGroupsMeet>0). The effect of increasing the
probability of encounters between groups with past histories was slightly positive but
asymptotic (Fig. 15). The effect of increasing the probability that the groups meeting and
joining would be unrelated was also positive and asymptotic, with a marked decline in
the proportion of whales hunting alone (Fig. 15). The increasing standard deviation as
simulations progressed across increasing ProbGroupsMeet, then ProbJoinRandomGroup
(Fig. 15) was an artifact of the higher variation in population size…i.e., increasing mean
group size to the optimum group size had the effect of increasing killing efficiency and
raising population size and variability. Group size was greatest during periods of increase
24
and smallest during population declines, leading to greater variability in mean group size
as an artifact of more variable population size. Adjusting encounter rates between
0
1
2
3
4
5
0 0.5 1Probability
Gro
up S
ize
ProbGroupsMeet
ProbJoinRandomGroup
0
0.05
0.1
0.15
0.2
0.25
0.3
1 2 3 4 5 6 7 8 9GroupSize
Perc
en
t F
reque
ncy
ProbJoinRandomGroup=0.1
ProbJoinRandomGroup=0.5
ProbJoinRandomGroup=0.9
Figure 15. For an optimum hunting group size of 3, parameters controlling
the daily probability of meeting and considering joining another group of
killer whales (ProbGroupsMeet), and the probability that it would be a
random group or a group of relatives (ProbJoinRandomGroup) had
positive effects on the mean group size of hunting killer whales (left). The
increase in mean group size with increasing ProbJoinRandomGroup was
accompanied by a marked decline in the proportion of whales hunting
alone (right).
predators and prey to control this effect showed the effect of ProbGroupsMeet and
ProbJoinRandomGroup on mean group size was robust. The mean group size counting all
adults and juveniles was greater than the optimum based on “adult equivalents” when
these controlling parameters allowed the greatest model flexibility in joining groups,
which was consistent with our expectations.
Model Validation
Model validity was assessed by concurrence of emergent population-level
properties of the model killer whales with comparable properties reported in the
literature. Specifically, maximum growth rates, age-sex composition, age-specific
survival, pregnancy and calving rates (Olesiuk et al. 1992), prey consumption rates and
the size of hunting groups (Baird and Dill 1995) were compared to values reported in the
literature. We also attempted to evaluate the plausibility of model behavior in conditions
of prey abundance and scarcity by comparison with other species of large mammals that
have been reported in those conditions (e.g., the demography of irruptive ungulate
populations or cyclic lynx (Lynx canadensis) suggest plausible demographic changes in
response to food abundance/scarcity).
Population Dynamics of Model Killer Whales
A representative simulation of killer whales preying on a superabundant population of
harbor seals was analyzed to evaluate whether killer whale population dynamics
conformed to those found in resident killer whales (Olesiuk et al. 1991) and large
mammals in general (Eberhardt 2002). There are no demographic data from field
populations of killer whales that can encompass the period and the range of growth rates
simulated. Olesiuk et al. (1990) estimated a life table of resident killer whales with a
25
stable population growth rate of λ=1.029. A stable projection from that table produced
similar age-class composition and survival rates to those of the initial growth phase in our
agent-based simulations (Table 2).
Table 2. Comparison of age structure and survival from agent-based simulations (SD in
parentheses) during 50 years of growth to those of a stable population estimated from a
life table of resident killer whales (Olesiuk et al. 1990) growing at a comparable rate.
Stable Projection Agent-Based Simulation
Growth Rate (λ) 1.029 1.032
Females >10 Years 0.40 0.41 (0.03)
Males >10 Years 0.24 0.19 (0.02)
Juveniles 1-10 Years 0.30 0.34 (0.03)
Calves 0.04 0.05(0.02)
Adult Female Survival 0.989 0.987 (0.020)
Adult Male Survival 0.969 0.966 (0.051)
Juvenile Survival 0.952 0.967 (0.037)
Calf Survival 0.96 0.895 (0.152)
Calving Rate (10-40) 0.14 0.18 (0.07)
The trajectory of killer whale and harbor seal abundance in a typical simulation
where both populations fluctuate are shown in Figure 16. From an initial population of 50
the killer whale population grew at a rate of just over λ=1.03 in the first 50 years,
reaching a peak of 280 after 66 years and fluctuating between 135 and 270 for the
remainder of the 1000-year simulation. Periods of decline were generally marked by
reduced calving rates and juvenile survival in comparison to periods of increase, leading
to poor recruitment (Fig. 16). Population trends in model killer whales were therefore
driven primarily by changes in calving rates and juvenile recruitment, with large
fluctuations in age structure that persisted for decades. This is consistent with the high
stability of adult survival and the species’ extreme longevity, but unique to large
predators. It is of interest that in this extremely long-lived species with a long post-
reproductive phase for females, modeled fluctuations in numbers were accompanied by
large shifts in population age-sex structure that affected reproductive potential. During
periods of decline, post-reproductive females came to outnumber reproductive ones, and
juveniles were reduced to less than half their proportion during periods of population
growth (Fig. 16). These features could be expected to lead to substantial lags in predator
numeric response to prey abundance, and unstable predator-prey interactions on long
time scales. This was observed in the illustrated 1000-year time series (Fig. 16), where
the mean lag between clear troughs in prey and predator numbers was 31 years (n=5,
range 16-38).
Consumption-Dependent Vital Rates
If we assume that most density-dependent changes in vital rates (Eberhardt and
Siniff 1977, Gaillard et al. 1998, Eberhardt 2002) are driven by consumption, what are
often thought of as density-dependent responses in vital rates are more usefully analyzed
as consumption-dependent responses in vital rates that control predator abundance. We
expected these to adhere to density-dependent patterns in that juvenile survival and
26
0
100
200
300
0 500 1000
Year
Kill
er
Wh
ale
s
0
500,000
1,000,000
1,500,000
Ha
rbo
r S
ea
ls
Killer Whales
Harbor Seals
0
100
200
300
0 100 200
Year
Kill
er
Wh
ale
s
0
0.2
0.4
0.6
0.8
1
Ra
te
Killer Whales Calving RateAbortion Rate Adult SurvivalJuvenile Survival Calf Survival
0
100
200
300
0 100 200Year
Kill
er
Wha
les
0
0.2
0.4
0.6
Pro
port
ion
Killer Whales JuvenilesAdult Females Adult MalesReproductive Females Post-reproductives
0
100
200
300
0 100 200
Year
Kill
er
Wh
ale
s
0
1
2
3
4
Gro
up
Siz
e o
r S
ea
ls K
ille
d
Killer WhalesSeals Killed/WhaleMean Group Size
Figure 16. A typical simulation of transient killer whales and a
superabundant, single prey population of harbor seals shows fluctuating
population size of predators and prey (top left) over 1000 years. Changing
vital rates (top right; rates are 5-year running averages), age structure
(bottom left), mean hunting group size and annual consumption rates
(bottom right) are illustrated for the first 200 years. For analyses,
juveniles are those 1-10 years old, and reproductive females include ages
10-40 years. Calving and calf survival rates are calculated as if data were
collected in summer (sensu Olesiuk et al. 1990).
reproductive rates should be the most responsive to changes in the per capita prey
consumption rates of killer whales. This expectation was confirmed by a high degree of
consumption-dependence in calving rate (as driven by the rate of abortions, Fig. 16) and
juvenile (especially calf and yearling) survival (P<<0.05, Fig. 17). These relationships
drove the strong shifts in juvenile recruitment and changing age structure apparent in the
trajectories of Fig. 16.
Surprisingly, finite growth rate (λ) was negatively correlated to total per capita
consumption rate (Fig. 17). This counter-intuitive result was driven primarily by changes
in population age structure as the population fluctuated in size (Fig. 17). Highest
consumption rates occurred when juvenile recruitment was low, leading to a high
proportion of adults whose larger body size required greater energy. This resulted in high
rates of per capita consumption while reproductive potential and population growth rate
were relatively low, and senescent mortality was increasing. This is an intriguing aspect
of the predator-prey relationship for transient killer whales that suggests caution when
27
0
0.2
0.4
0.6
0.8
1
1.15 1.20 1.25 1.30 1.35 1.40
Seals Eaten/Adult Whale/Day
Ann
ua
l R
ate
Pregnancy
Calving
Abortion
0.4
0.6
0.8
1
0.00 0.50 1.00 1.50
Seals Eaten/Whale/Day
An
nu
al S
urv
iva
l
Adults
Juveniles
Yearlings
Calves
0.8
0.9
1.0
1.1
0.9 1.0 1.1 1.2Seals Killed/Whale/Day
Pop
ula
tion
Gro
wth
(λ
)
0
50
100
150
200
250
300
0 50 100 150 200Year
Kill
er
Wh
ale
s
0.0
0.1
0.2
0.3
0.4
0.5
Pro
po
rtio
n
Killer Whales
Juveniles
Figure 17. Scatterplots from the first 200 years of a simulation of killer
whales preying on a single, superabundant prey population of harbor seals.
Calving rates were positively and abortion rates negatively correlated with
the number of seals eaten/adult (>10 years) killer whale/day (top left;
reproductive females were defined as those from 10-40 years of age).
Annual survival rates were positively correlated with class-specific
consumption (top right), but finite population growth (λ) was negatively
correlated with total per capita killing rate (bottom left) due to shifting age
structure during population fluctuations (bottom right, see text).
examining killer whale functional or numeric responses to prey abundance. Functional
and numeric responses of killer whales were, at best, weakly correlated to both seal
abundance and ratio of seals/killer whales when considered without time lags. When
time lags were considered using cross correlations, the strongest responses were to seal
abundance directly rather than to seals available per killer whale {i.e., model killer
whales were more nearly “prey-dependent” than “ratio dependent” (Arditi and Ginzburg
1989)}. Seal abundance was positively correlated (r = 0.395) to killing rate per killer
whale 24 years earlier, and negatively correlated ((r = 0.390) to killing rate 19 years later
(Fig. 18). Similarly, killer whale numeric response (λ) was most highly correlated (r = -
0.39) with seal abundance 13 years earlier (Fig. 18). Such simulations suggest that the
standing age and social structure, with the broad range of age-specific body sizes and
reproductive potentials possible with killer whales, are more important to interpreting
killer whale impact on their prey, and predator-prey dynamics in general than killer whale
numbers per se.
28
0.90
0.95
1.00
1.05
1.10
1.15
1.20
1100000 1300000 1500000
Seals Available in Year-19
Se
als
Kill
ed/W
hale
/Da
y
0.80
0.85
0.90
0.95
1.00
1.05
1.10
1100000 1300000 1500000
Seals Available in Year-13
Gro
wth
Rate
(λ)
Figure 18. Functional (left) and numeric (right) responses of killer whales
to prey abundance were more dependent on absolute abundance than the
seals/killer whales ratio, with anti-regulatory properties indicated by long
time lags.
Multiple Prey and Other Features
We used the single-prey model as a baseline to explore interactions between killer
whales and multiple prey by calculating the amount of biomass provided by the addition
of a new prey species to the model (number of prey × expected encounter rate × average
vulnerability), and reducing the number of harbor seals to remove a comparable biomass
of those prey from the whales’ diet. In this way the addition of new prey species would
lead to a similar standing stock of killer whales. Two species of particular interest are
Steller sea lions, because of their threatened and endangered status in parts of their range,
and gray whales because of their apparent importance as prey during a narrow window of
time in the spring when calves are known to be preyed upon. We added Steller sea lions
and gray whales at stock sizes similar to those existing on the North American coast from
California to SE Alaska (Appendix), and reduced the super-abundant population of
harbor seals to compensate for the two additional prey. Steller sea lions were assumed to
be somewhat less vulnerable than harbor seals, and to require slightly larger group sizes
for optimal foraging. Gray whales were the least vulnerable and required optimal group
sizes of 5 or 10 adult killer whales to reach optimum hunting efficiency on calves and
adults, respectively. Seasonal availability and increased hunting specialization on gray
whale calves was simulated by limiting the window of time during each year to days 50-
150, and increasing the encounter rate for them in comparison to harbor seals and Steller
sea lions.
The 3-species simulation is illustrated with a period of killer whale population
growth and equilibration with the prey community followed with a “regime shift”
affecting the primary prey population of harbor seals, a return to the original regime, and
later by the harvest removal of seasonally available gray whales. We reduced the
carrying capacity of harbor seals in year 200 by ~60% for a period of 30 years, then
returned the system to its starting parameters until year 400, when 12,000 gray whales
were removed by harvest over 10 years. The simulation (Fig. 19) shows the increase in
killer whales and the reduction of prey populations as killer whale numbers increased.
The impact of the growing killer whale population on gray whales was small, while the
fluctuations in both pinniped populations are roughly synchronous. As modeled, harbor
seals represent a large prey base (that might simulate a conglomeration of multiple
29
species) that dominates the effect on killer whale population dynamics. Their sudden
precipitous decline in year 200 caused a decline in killer whale numbers that reached its
nadir 59 years after the first regime shift and 29 years after the carrying capacity for
harbor seals had returned to its original condition (Fig. 19). This was caused primarily by
the collapse in killer whale recruitment for most of that 30 year period of poor harbor seal
conditions. A small increase (~15%) in predation on the other 2 species occurred in the
decade following the harbor seal reduction, but both species increased their numbers as
the killer whale population declined. The effect on killer whale numbers and recruitment
of removing gray whales in years 400-409 was not noticeable. Mean group size of killer
whales increased from 2.73 to 3.66 when gray whales became available seasonally.
0
100
200
300
0 200 400 600
Year
Kill
er
Whale
s
0
0.2
0.4
0.6
0.8
1
Pro
po
rtio
n
Killer WhalesJuvenilesReproductive Females
0
400000
800000
1200000
1600000
0 200 400 600
Year
Ha
rbo
r S
eals
0
10000
20000
30000
40000
50000
60000
Se
a L
ion
s &
Gra
y W
ha
les
Harbor Seals
Steller Sea Lions
Gray Whales
Figure 19. Simulation of a 3-species prey community (see text) over 600
years in which a 30-year “regime shift” reducing the carrying capacity of
harbor seals by ~60% begins in year 200, and a harvest of 12,000 gray
whales occurs in years 400-409.
Discussion
Our overall objective of producing an individual-, or agent-based model with
emergent properties similar to those that have been estimated from live killer whales was
met, and some non-intuitive properties of those models were discovered. Rates of prey
consumption were (not surprisingly, given their shared derivation) similar to those
published (Barrett-Lennard et al. 1995, Baird and Dill 1996, Williams et al. 2004),
though somewhat higher due to inclusion of growth and reproductive costs.
Demographically, baseline rates were easily adjusted to mimic those expected under
exponential growth, while energetic mechanisms provided plausible responses in vital
rates when prey became scarce. Confirmation that such changes occur in wild
populations is not now feasible, but the use of an individual-based model might identify
other indirect properties that could strengthen an argument that food shortages affect
demography of transient killer whales. For example, the existence of thresholds in body
mass that lead to decreasing lactation, abortion or starvation might be confirmed and
estimated as potential physiological mechanisms that would logically affect demography;
field studies would be unlikely to obtain sufficient sample sizes to definitively link
consumption to these mechanisms.
Many other results, particularly those pertaining to long-term predator-prey
dynamics, are speculative, but rich in detail about possible mechanisms and
consequences. There are essentially no theoretical models relevant to a single predator
30
and the number of prey routinely taken by killer whales, or of a predator with such
longevity and likely demographic inertia….factors that lead to substantial lags in the prey
scenarios modeled here. The importance of standing age structure to the interpretation of
demographic trends in killer whales was amply demonstrated in model simulations. In
some cases the prey population began its recovery over 30 years before the killer whale
population, and numbers of killer whales sometimes remained stable for decades into a
prey decline before dropping. The ability to determine age- and sex-specific
demographic rates and age structure in killer whale field studies is essential to
understanding population productivity and other potential responses to changing prey
availability. This is due to the narrow age window for reproduction and the existence of a
large post-reproductive class in killer whale populations. The predator-prey dynamics
that follow from this do not lend themselves to classical forms of difference equation
models, though the output from individual-based simulations may aid in developing
simpler mathematical models with long time lags.
Current controversies about the role of transient killer whales in the decline of
species such as Steller sea lions have taken little account of the reciprocal numeric effect
such declines might have on killer whales. The simulations performed here indicate that
such affects could be profound, and suggest some attendant changes that might be
relevant to future studies. One is that there are significant lags in total numbers of killer
whales following prey reduction: reduction of prey numbers might increase the impact of
killer whale predation and accelerate a prey decline until the effects of reduced
recruitment and shifting age structure act to decrease killer whale numbers. The practical
significance of this is that reductions in prey could have significant effects on the age-sex
structure of transient killer whales that are persistent and measurable with current
methods at the population or social group level, even while total numbers of killer whales
appear stable. In simulations per capita consumption rate varied with age structure and
reproductive success, which also could affect assessments of the impact of transient killer
whales on prey populations. Estimates of individual metabolic requirements may need to
be weighted by the actual age-sex structure, not simple estimates of average killer whales
(Williams et al. 2004) or a calculated stable structure (Barrett-Lennard et al. 1995). The
potential to explore these relationships with the present model has not been exhaustively
pursued here.
Future Direction
Our intention was to implement an individual-based model built on biological
components that have empirical support, and to explore the emergent properties of
models based on these biologically realistic elements. Uncertainties about the likely value
of many parameters are important to those properties, but assumptions about the model
structure also need to be addressed in the future. We feel that the most obvious and
important of these is the simplifying assumption that hunting groups of killer whales and
their prey encounter each other at random, without spatial structure and with very little
temporal structure. The various species on which transient killer whales prey have widely
different dispersion across the range of a given killer whale stock, as well as pronounced
differences in habitat use. Strictly speaking, it is impossible for a killer whale group to
daily traverse a small part of its range and have the same probability of encountering, say,
a Steller sea lion and Pacific white-sided dolphin every day of the year. Basing the
predator-prey interaction on established theoretical assumptions allowed both for useful
31
comparisons to conclusions from theoretical ecology, and for segue into spatial models in
the next generation of a killer whale IBM. The natural spatial implementation would be a
spatial grid (hypothetical or one based on GIS maps) which incorporates information on
the dispersion of prey stocks. The probabilistic mechanisms of encounter between groups
of transient killer whales and between those groups and prey are much more plausible at
smaller spatial scales, and a variety of decision mechanisms could be implemented and
tested against the movement patterns inferred from resighting data or satellite telemetry.
Spatial segregation of prey is a plausible mechanism providing potential for refuges and
aggregation for prey species, factors that tend to prevent prey extinction and increase
model stability in simpler predator-prey models (Hassell 1981, Akcakaya 1992,
McCauley et al. 1993).
Conclusion
There is no intention to imply specific predictive ability to the model described
here. Its value is primarily heuristic and the lessons are general. Mechanisms of
interaction between transient killer whales and their prey can be simulated and altered,
and the logical consequences of such changes in our conceptualization can be observed.
The model often suggests responses to real changes that we might not at first predict,
suggesting measurements we could make in the field to detect demographic or ecosystem
changes. Changes in age structure and extreme lags in predator-prey dynamics are two
such responses to declining prey abundance suggested by the model. To the extent that
energetic requirements of transient killer whales are adequately captured in the model, we
might expect that changing consumption patterns might cause changing demographic
patterns, and simulations can suggest what data to collect in order to detect these
responses. Our experience in developing and experimenting with these models has been a
positive one of being forced to examine the details of physiological and ecological
processes from their first assumptions to logical consequences that follow. Even where
realism is stretched to breaking, we believe that our understanding of how real systems
work can only be improved by the exercise of modeling. Our primary hope is that
students of killer whale ecology will find a useful tool herein, and be stimulated to use
and improve on this beginning.
Acknowledgments
We would like to thank Craig Matkin, Lance Barrett-Lennard, Paul Wade, John
Durban, Marilyn Dahlheim and Don Siniff for helpful conversations during the creation
of the model, and absolve them of any responsibility for its shortcomings. We also thank
the Marine Mammal Commission (http://www.mmc.gov) for funding student support for
this project, the National Marine Mammal Laboratory for support to JWT and University
of Alaska Anchorage for support to KJM.
Literature Cited
Akcakaya, H. R. 1992. Population cycles of mammals: evidence for a ratio-dependent
predation hypothesis. Ecological Monographs 62:119-142.
Arditi, R., and L. R. Ginzburg. 1989. Coupling in predator-prey dynamics: ratio-
dependence. Journal of Theoretical Biology 139:311-326.
32
Aviles, L., P. Abbot, and A. D. Cutter. 2002. Population ecology, nonlinear dynamics,
and social evolution. I. associations among nonrelatives. The American Naturalist
159:115-127.
Baird, A. W., and L. M. Dill. 1995. Occurrence and behaviour of transient killer whales:
seasonal and pod-specific variability, foraging behaviour, and prey handling.
Canadian Journal of Zoology 73:1300-1311.
Baird, R. W., and H. Whitehead. 2000. Social organization of mammal-eating killer
whales: group stability and dispersal patterns. Canadian Journal of Zoology
78:2096-2105.
Baird, W. R., and M. L. Dill. 1996. Ecological and social determinants of group size in
transient killer whales. Behavioral Ecology 7:408-416.
Barlow, J., and P. L. Boveng. 1991. Modeling age-specific mortality for marine mammal
populations. Marine Mammal Science 7:50-65.
Barlow, J., and P. Clapham. 1997. A new birth-interval approach to estimating
demographic parameters of humpback whales. Ecology 78:535-546.
Barrett-Lennard, L. G., K. Heise, E. Saulitus, G. Ellis, and C. Matkin. 1995. The impact
of killer whale predation on Steller sea lion populations in British Columbia and
Alaska. North Pacific Universities Marine Mammal Research Consortium,
Vancouver, British Columbia.
DeMaster, D. P., A. W. Trites, P. Clapham, S. Mizroch, P. Wade, R. J. Small, and J. Ver
Hoef. 2006. The sequential megafaunal collapse hypothesis: testing with existing
data. Progress in Oceanography 68:329-342.
Eberhardt, L. L. 2002. A paradigm for population analysis of long-lived vertebrates.
Ecology 83:2841-2854.
Eberhardt, L. L., and D. B. Siniff. 1977. Population dynamics and marine mammal
management policies. Journal of the Fisheries Research Board of Canada 34:183-
190.
Estes, J. A., M. T. Tinker, T. M. Williams, and D. F. Doak. 1998. Killer whale predation
on sea otters linking oceanic and nearshore ecosystems. Science 282:473-476.
Ferrero, R. C., and T. R. Walker. 1996. Age, growth, and reproductive patterns of the
Pacific white-sided dolphin (lagenorhynchus obliquidens) taken in high seas drift
nets in the Central North Pacific Ocean. Canadian Journal of Zoology 74:1673-
1687.
Ferrero, R. C., and W. A. Walker. 1999. Age, growth, and reproductive patterns of Dall's
porpoise (Phocoenoides dalli) in the central North Pacific Ocean. Marine
Mammal Science 15:273-313.
Ford, J. K. B., G. M. Ellis, D. R. Matkin, K. C. Balcomb, D. Briggs, and A. B. Morton.
2005. Killer whale attacks on minke whales: prey capture and antipredator tactics.
Marine Mammal Science 21:603-619.
Gaillard, J. M., M. Festa-Bianchet, and N. G. Yoccoz. 1998. Population dynamics of
large herbivores: variable recruitment with constant adult survival. Trends in
Ecology and Evolution 13:58-63.
Grimm, V., and S. F. Railsback. 2005. Individual-based modeling and ecology. Princeton
University Press, Princeton, New Jersey.
33
Hassell, M. P. 1981. Arthropod predator-prey systems. Pages 105-131 in R. M. May,
editor. Theoretical Ecology: Principles and Applications. Sinauer, Sunderland,
Massechusetts.
Heise, K., B.-L. L. G., E. Saulitis, C. Matkin, and D. Bain. 2003. Examining the evidence
for killer whale predation on Steller sea lions in British Columbia and Alaska.
Aquatic Mammals 29:325-334.
Heyning, J. E. 1988. Presence of solid food in a young calf killer whale (Orcinus orca).
Marine Mammal Science 4:68-71.
Holmes, E. E., and A. E. York. 2003. Using Age Structure to Detect Impacts on
Threatened Populations: a Case Study with Steller Sea Lions. Conservation
Biology 17:1794-1806.
Horwood, J. 1990. Biology and Exploitation of the Minke Whale. CRC Press, Boca
Raton, Florida.
Kriete, B. 1995. Bioenergetics in the killer whale, Orcinus orca. Ph.D. University of
British Columbia, Vancouver.
McCauley, E., W. G. Wilson, and A. M. De Roos. 1993. Dynamics of age-structured and
spatially structured predator-prey interactions: individual-based models and
population-level formulations. American Naturalist 142:412-442.
Metzgar, L. H., and E. Boyd. 1988. Stability properties in a model of forage-ungulate-
predator interactions. Natural Resource Modeling 3:3-43.
Mizroch, S. A., and D. W. Rice. 2006. Have North Pacific killer whales switched prey
species in response to depletion of the great whale populations? Marine Ecology
Progress Series 310:235-246.
Moen, R., Y. Cohen, and J. Pastor. 1998. Linking moose population and plant growth
models with a moose energetics model. Ecosystems 1:52-63.
Moen, R., J. Pastor, and Y. Cohen. 1997. A spatially explicit model of moose foraging
and energetics. Ecology 78:505-521.
North, M. J., N. T. Collier, and J. R. Vos. 2006. Experiences Creating Three
Implementations of the Repast Agent Modeling Toolkit. ACM Transactions on
Modeling and Computer Simulation 16:1-25.
Olesiuk, P. F., M. A. Bigg, and G. M. Ellis. 1990. Life history and population dynamics
of resident killer whales (Orcinus orca) in the coastal waters of British Columbia
and Washington State. Report of the International Whaling Commission Special
Issue 12:209-243.
Parrish, J. K., S. V. Viscido, and D. Grunbaum. 2002. Self-organized fish schools: an
examination of emergent properties. Biol. Bull. 202:296-305.
Rice, D. W. 1978. The humpback whale in the North Pacific: distribution, exploitation
and numbers. Pages 29-44 in. U.S. Marine Mammal Commission, Washington,
DC.
Springer, A. M., J. A. Estes, G. B. van Vliet, T. M. Williams, D. F. Doak, E. M. Danner,
K. A. Forney, and B. Pfister. 2003. Sequential megafaunal collapse in the North
Pacific Ocean: an ongoing legacy of industrial whaling? Proceedings of the
National Academy of Sciences 100:12223-12228.
34
Steiger, G. H., and J. Calambokidis. 2005. Killer whale rake mark scarring on humpback
whale flukes: a demographic and temporal analysis on the California-Oregon-
Washington feeding aggregation (abstract). Pages 267 in 16th Bienial Conference
on the Biology of Marine Mammals. Society for Marine Mammalogy, San Diego,
California.
Trites, A. W., V. B. Deeke, E. J. Gregr, J. K. B. Ford, and P. F. Olesiuk. 2007 In Press.
Killer whales, whaling and sequential megafaunal collapse in the North Pacific: a
comparative analysis of the dynamics of marine mammals in Alaska and British
Columbia following commercial whaling. Marine Mammal Science 23:000-000.
von Bertalanffy, L. 1938. A quantitative theory of organic growth (Inquiries on growth
laws. II). Human Biology 10:181-213.
Wade, P. R. 2002. A Bayesian stock assessment of the eastern Pacific gray whale using
abundance and harvest data from 1967-1996. Journal of Cetacean Research and
Management 4:85-98.
Wade, P. R., L. G. Barrett-Lennard, N. A. Black, V. N. Burkanov, A. M. Burdin, J.
Calambokidis, S. Cerchio, M. E. Dahlheim, J. K. B. Ford, N. A. Friday, L. W.
Fritz, J. K. Jacobsen, T. R. Loughlin, C. O. Matkin, D. R. Matkin, S. M.
McCluskey, A. V. Mehta, S. A. Mizroch, M. M. Muto, D. W. Rice, R. J. Small, J.
M. Straley, G. R. Van Blaricom, and P. J. Clapham. 2007 In Press. Killer whales
and marine mammal trends in the North Pacific-a re-examination of evidence for
sequenetial megafauna collapse and prey-switching hypothesis. Marine Mammal
Science 23:000-000.
Williams, T. D., J. A. Estes, D. F. Doak, and A. M. Springer. 2004. Killer appetites:
assessing the role of predators in ecological communities. Ecology 85:3373-3384.
Winship, A. J., A. W. Trites, and D. A. S. Rosen. 2002. A bioenergetic model for
estimating the food requirements of Steller sea lions Eumetopias jubatus in
Alaska, USA. Marine Ecology Progress Series 220:291-312.
35
Appendix: Prey species and associated parameters
Sea Otters – Wade et al. (2007 In Press) estimated population size of sea otters from SE
Alaska to California at just under 18,000. With relative stability in California, and rate of
expansion apparently slowed in Alaska and BC, we assumed the population was above
MNPL, with growth slowing from rmax. Life history parameters provided by Gerber et al.
were used to set vital rates that produced rmax=0.18. Due to a greater diversity of
observed vital rates and population growth rates in relation to density, a slightly more
conservative approach to equilibrium than that used for remaining species was assumed,
with MNP ~0.65.
While evidence supports substantial impact on sea otters by transient killer whales
in western Alaska, there is little evidence for significant predation by transient killer
whales in SE Alaska or further south, so default vulnerability maxima were set very low
for both pups and non-pups, with little increase in success with increasing group size
(http://www.math.uaa.alaska.edu/~orca/).
Harbor Seals – The stocks on the west coast of North America were combined by Wade
et al. (2007 In Press) to estimate 102,657 available to transient killer whales. Observed
rates of population growth by small populations of harbor seals are variable, so we
assumed the observed estimate was >80% of equilibrium. We assumed rmax = 0.12 and
the parameters to produce that rate given in Http://www.math.uaa.alaska.edu/~orca/.
Density dependent parameters were selected to produce MNPL between 75-80% of
equilibrium, with greatest changes in calf survival, reproduction and adult survival,
respectively.
Harbor seals are widely reported as prey to transient killer whales, and dominate
the diet in British Columbia (Baird and Dill 1995). In that study, pups were more
abundant in the diet than adults, and the relationship of hunting success to group size
suggested an optimum group size of 3 across all prey types (small cetaceans and sea lions
were also taken). We modeled vulnerability to reproduce these conclusions, with
maximum vulnerability reached at smaller group size for harbor seal pups.
California Sea Lions – The 2001 estimate based on corrected pup counts was ~240,000
animals, with recent growth of 0.06 annually, though El Nino conditions and recent
disease outbreaks suggest the population is close to equilibrium. Default parameter values
produce an equilibrium close to the 2001 estimate and rmax=0.12.
Default vulnerability parameters are the same as for harbor seals, but encounter
rates are lower due to the more restricted range and concentration around a small number
of rookeries. Pups are born in May and June, but the model uses July 1 as birthdate (day
of year) to reflect the approximate time that pups regularly enter the water, becoming
potentially vulnerable to predation by killer whales.
Steller Sea Lions – The eastern stock extends from SE AK to California, likely
numbering around 43,000 animals when rookery and haul-out counts are corrected for
animals at sea(Wade et al. 2007 In Press). Pups born in SE Alaska have increased at a
rate of r=0.059 from 1979-2000 but this rate had declined in the later part of that period,
and our default population for the eastern stock is modeled for rmax = 0.10. Vital rates
36
were derived from an equilibrium life table provided by (Holmes and York 2003). While
rookeries in Oregon and California are relatively stable, pups born annually at 3 rookeries
in SE Alaska have been increasing at roughly r=0.059, though that rate has declined in
recent years. Rookeries in British Columbia are also expanding after prolonged culling
and extirpation of rookeries in the mid-20th
century.
Default vulnerability parameters assume that the difference between pups and
non-pups in maximum vulnerability is greater than that of harbor seals, due to the large
degree of sexual dimorphism and average size of adults, many of which are adult males
that pose a significant risk of injury to killer whales trying to eat them. Pups are born in
May and June, but the model uses July 1 as birthdate (day of year) to reflect the
approximate time that pups regularly enter the water, becoming potentially vulnerable to
predation by killer whales.
Northern Elephant Seals – The U.S. population was derived from a remnant population
of a 10’s or hundreds of seals surviving commercial hunting in the 19th
century. The first
pup was born on Ano Nuevo island in 1961, but the present population occupies 3 islands
some mainland beaches, growing exponentially at rmax = 0.078. The equilibrium level of
this stock is unknown, but the continued growth and expansion to new breeding beaches
suggests it could be some way off. We assume that the population is starting to slow its
growth and will level off around 145,000.
Birthing date and mass of pups is adjusted to early April to account for delayed entry into
the ocean and availability to killer whales. First year growth in mass is assumed to be
minimal, as pups enter the water with huge fat reserves and end the year as much leaner,
longer seals of similar mass. Assumed mass of non-pups is much less than adults to
account for the likely greater vulnerability of juveniles and females to killer whale
predation. Vulnerability as a function of killer whale group size was similar to that of
Steller sea lions, but encounter rates were reduced to reflect more restricted range and
seasonal availability.
Harbor Porpoise – (Barlow and Boveng 1991) suggested a maximum theoretical growth
rate of 0.094 for harbor porpoises, but actual estimates of porpoise and dolphin growth
rates have been well below this. The default parameters used here reflect maximum age
of less than 15 years and annual reproduction, producing a growth rate rmax=0.05
Harbor porpoise are of similar size to harbor seals, but faster and swimmers
requiring energetic pursuit by killer whales. Optimal group sizes for capture of calves and
non-calves were assumed similar to harbor seals, but maximum vulnerability was
assumed to be lower.
Dall’s Porpoise – Excellent information on body growth and reproductive parameters
were obtained from (Ferrero and Walker 1999). A theoretical maximum growth rate of
0.12 was assumed. Capturing Dall’ Porpoises requires energetic pursuit by killer whales,
but yields less success than for smaller harbor porpoises. Also, successful group size was
more variable across these two species than across pinniped species observed by
Dahlheim and White (Pers. comm.). Default parameters assumed that optimum group
size was slightly higher, and maximum vulnerability slightly lower than for harbor
porpoise. Optimum group size was still below that observed when transients killed gray
37
whales (below). This means that vulnerability was assumed to increase more slowly with
group size than for harbor seals and harbor porpoises.
Pacific White-sided Dolphin – Wade et al. (2007 In Press) estimated a population size of
59,274 available on the west coast, and we assumed that this was >80% of equilibrium.
Estimates of reproductive rates (Ferrero and Walker 1996) and the maximum known ages
of around 40 years suggest vital rates more similar to Spotted Dolphins than to the
similar-sized Dall’s porpoises. We assumed rmax = 0.05 with lower reproductive and
mortality rates than used in Dall’s porpoises (http://www.math.uaa.alaska.edu/~orca/).
Density dependent parameters were selected to produce MSP between 75-80% of
equilibrium, with greatest changes in calf survival, reproduction and adult survival,
respectively (http://www.math.uaa.alaska.edu/~orca/). Vulnerability was assumed similar
to that of Dall’s porpoises.
Minke Whales – The Pacific stock of minke whales is the most poorly known of the large
whales considered in this model, but they are also the smallest and most vulnerable to
attack by killer whales (Ford et al. 2005), with a higher proportion of adult kills among
those observed than in gray whales. It was therefore essential to include them as potential
prey. Population size along the west coast is estimated at ~1015 (Wade et al.2007 In
Press). Minke whales were not a target of commercial whaling in the 20th
century so this
might be considered an equilibrium level. It is, however, a much smaller density than
other populations in the Atlantic and Antarctic, and predation by killer whales may be a
significant limiting factor. Therefore, the present level is unlikely to be a purely
density-dependent equilibrium in the absence of predation, but there is no way to derive
an estimate for a purely density-dependent minke whale population. We therefore chose
to assume the present level to be below, but near the MSP level (80% of equilibrium) to
make it as robust to predation pressure as possible. Density dependent parameters were
selected to produce MSP between 75-80% of equilibrium, with greatest changes in calf
survival, reproduction and adult survival, respectively. We used the highest values for
adult survival and fecundity provided by(Horwood 1990), based primarily on commercial
catches in Antarctica and the North Atlantic. Assuming minimum juvenile mortality was
~ twice that of adults, rmax = 0.088. Density dependent parameters were selected to
produce MSP between 75-80% of equilibrium, with greatest changes in calf survival,
reproduction and adult survival, respectively (Http://www.math.uaa.alaska.edu/~orca/).
There are no estimates of minke whale vulnerability except anecdotes that
indicate adults are more vulnerable than gray whale adults. They are capable of escaping
killer whale attacks during long chases, but not if trapped by geography (Ford et al.
2005). They are also smaller, so we assumed that maximum vulnerability would be
higher and be attained by smaller groups of killer whales than is assumed for gray whales
(below). Minke whale body weights available for consumption by killer whales were
derived from parameters for length and age, and weight at length equations (Horwood
1990). Non-calf mass was the mean of mass at mid-year age weighted by the expected
age structure. Fat content of minke whales is the lowest of any large whale and set at
2.35 kcals/gm wet weight (Horwood 1990).
38
Gray Whales – The western pacific gray whale stock is estimated to have grown at
0.024-0.044 from 1967-1998, reaching a maximum population estimate of ~26,000 in
1998. Gray whale numbers subsequently declined after environmental changes in their
feeding areas, and with known predation on calves. We used 25000 to be the default
equilibrium in our model. Birth intervals are longer than those of humpback whales that
have been observed to increase r = 0.08. Maximum survival estimates are high (Wade
2002). We used default vital rates that produced rmax = 0.05 rather than the observed rate
of growth in the late 20th
century because of the likelihood that observed rates were
obtained while killer whale predation was occurring. Density dependent parameters were
selected to produce MSP between 75-80% of equilibrium, with greatest changes in calf
survival, reproduction and adult survival, respectively.
Predation by transient killer whales on gray whales is known to occur primarily
on calves, and group size of attacking killer whale is over twice that during attacks on
pinnipeds (Dalhleim and White, Pers. comm.). We therefore modeled vulnerability to
increase with group size more slowly than for pinniped prey, producing a maximum rate
of return for individual killer whales at group sizes ~ 8 for calves, and 13 for non-calves,
with maximum vulnerability of calves being 4 times that of adults (0.1).
Humpback Whales – Two stocks overlap the population of west coast transients. Wade et
al. used estimates of the eastern North Pacific stock together with that portion of the
Central North Pacific stock that occupies SE Alaska in summer to conclude that ~2352
humpbacks were available to west coast transients. For the entire North Pacific stock,
(Rice 1978) estimated that ~ 15,000 humpbacks were present prior to commercial
whaling, which is roughly twice the present NOAA stock assessment. Therefore, the
default equilibrium level used in the model population was ~5,000.
Maximum rate of growth of the population has been reported as r=0.08. (Barlow
and Clapham 1997) provided parameter estimates suitable to derive maximum birth and
survival for the Atlantic population that was growing at 0.065, which were adjusted
slightly to obtain default values that produced the 8% growth rate observed in the Pacific
(http://www.nmfs.noaa.gov/prot_res/PR2/Stock_Assessment_Program/individual_sars.ht
ml). Density dependent parameters were selected to produce MSP between 75-80% of
equilibrium, with greatest changes in calf survival, reproduction and adult survival,
respectively (http://www.math.uaa.alaska.edu/~orca/).
Evidence of predation is seen almost entirely on calves, usually before the spring
migration in the E North Pacific stock, with roughly 11% seen with rake marks and 7%
acquiring scars after first being seen (Steiger and Calambokidis 2005). Other reports of
large whales being attacked involved large numbers of killer whales. Prey vulnerability
was therefore considered to be a more gradually increasing function of killer whale group
size than other prey, with calf vulnerability greatly exceeding that of non-calves. Non-
calf vulnerability is likely to be primarily among juveniles in this and other large whales.
Little is known about predation rates on any of the larger whales, but population
characteristics and vulnerability to predation may be similar amongst them. Also, a single
large whale is likely to satiate any group of killer whales when killed. Therefore, the
modeled humpback population might be used as a surrogate for the remaining large
whales by simply increasing the modeled stock size and equilibrium levels.